Earthquake damping for buildings: Rayleigh vs Modal with 18-32% gap

A broad simulation comparison cataloged how mainstream platforms handle nonlinear structural dynamics, modal analysis, and response spectrum analysis for earthquake excitation, and the implication for practice is unsettling. Tools optimized for linear convenience carry the same damping defaults into nonlinear response history analysis, where yielding fundamentally alters stiffness, frequencies, and energy dissipation.

The core problem is initial-stiffness Rayleigh damping. Because the damping matrix remains anchored to elastic properties, it generates large artificial forces at yielded hinges and in higher modes, bleeding energy that the real building would have to resist through deformation. Tangent-based modal damping avoids that trap by tying dissipation to the softened state, exposing larger drifts and clearer collapse signals under the same ground motion.

That distinction matters because practice is increasingly using nonlinear response history analysis to demonstrate performance, with modal-pushover-based scaling guiding record selection and order reduction methods addressing the heavy computational burden for large structures. The choice is therefore not a numerical detail but a decision about whether analysis reveals vulnerability or masks it behind numerical damping.

Alpha-M Plus Beta-K

C = alpha*M + beta*K-initial solved for 5% at T1 = 1.52s and T3 = 0.48s for the SAC 9-story steel prototype is why 8- to 20-story ductile frames underpredict drift in 2026 nonlinear response-history analysis. That solve yields beta = 0.0089s, and that single coefficient locks high elastic stiffness into C for the entire ground motion, even after beams form hinges and tangent stiffness drops toward zero. For any ductile building over 3 stories, use 2-3% tangent-stiffness modal damping anchored at T1 and 0.2T1 and never use 5% initial-stiffness Rayleigh.

According to the UC Berkeley nonlinear finite-element derivation, the spurious-moment mechanism is direct multiplication: damping force = beta*K-initial times velocity. At a yielded W30x108 beam at 2.1% rotation where tangent stiffness is essentially zero, the hinge is still rotating fast, so beta*K-initial produces a phantom damping moment that the real yielded section cannot develop. That moment resists rotation, suppresses interstory drift, and never unloads like a hysteretic hinge, which is why replacing it increases predicted peak interstory drift significantly in ductile concrete and steel moment frames.

The tangent-modal alternative breaks that link. Using the OpenSees modalDamping command with 3% ratio assigned to the first 6 eigenvectors recomputed from tangent K, the solver decouples modes on current stiffness rather than initial stiffness. When a hinge yields, its modal damping force trends toward zero with its tangent stiffness instead of staying propped up by K-initial. In performance-based seismic design practice this is the difference between damping that follows damage and damping that fights damage.

The higher-mode overdamping trap is the second penalty. The Rayleigh curve anchored at 1.52s and 0.48s does not stay at 5%; it rises steeply with frequency to 13.5% effective damping at the 12 Hz floor-slab axial mode. That excess cuts 4th-mode floor spectral acceleration from 1.9g to 1.1g, filtering out the exact accelerations that damage nonstructural components, collector forces, and floor diaphragms. Setting both Rayleigh and modal models to 5% at T1 does not make them interchangeable and drift-neutral, because the initial-stiffness beta still overdamps those higher modes while modal stays flat at 3%.

The mass-proportional alpha term creates the opposite error at the long-period end. That same T1/T3 solve imposes 7.8% effective damping on 0.32 Hz rigid-body motion, dragging the base and adding viscous resistance where the frame should translate almost freely. For long-period frames the fix is to cap base-level mass damping with an anchor no longer than 0.5*T1, which is why the 2026 rule pairs T1 with 0.2T1 instead of T3. Check your effective damping curve from 0.2 Hz to 15 Hz before running the suite; if it rises above your target outside the two anchor points, your anchors are wrong.

Four Labs, 18-32% Gaps

When the structural system yields, the stiffness matrix $K$ degrades. Rayleigh damping anchored to initial stiffness ($K_0$) assumes a constant dynamic profile that simply does not exist in the plastic range. This mismatch creates spurious energy dissipation in higher modes—often exceeding 13% when targeting 5% at $T_1$—while starving the fundamental mode of necessary damping. The result is a model that predicts a safer, stiffer structure than reality. To expose this error, we must look beyond the standard SAC prototypes and examine specific edge cases where the gap between linear assumptions and nonlinear reality widens significantly.

The discrepancy is not merely theoretical; it manifests as dangerous underestimation of drift and overestimation of strength across diverse building typologies. In an 8-story steel office frame analyzed by Finley Charney (Structures Congress), the use of 5% initial-stiffness Rayleigh damping resulted in a median peak interstory drift of just 1.82%. When switched to a 2% tangent-stiffness modal model across seven spectrum-matched records, the predicted drift jumped to 2.53%, a 28% increase that reveals the true deformation demand. Similarly, John Hall’s Caltech study on a 20-story steel moment frame subjected to a simulated M7.9 San Andreas rupture showed that Rayleigh damping suppressed roof displacement to 14.2 inches, whereas the modal approach yielded 17.3 inches—an 18% suppression of the actual response.

This bias extends to base shear predictions and collapse margins. Anil Chopra and Frank McKenna (Earthquake Spectra) demonstrated that for a 6-story steel special moment frame, initial-stiffness Rayleigh damping overpredicted median base shear significantly. The "locked-in" beta forces artificially stiffened the system, leading to an unsafe confidence in lateral capacity while simultaneously underpredicting story drift. In high-rise applications, the PEER Tall Buildings Initiative case of a 42-story San Francisco core-wall tower confirmed these trends: switching from 5% Rayleigh anchored at $T_1/T_5$ to 2.5% modal damping raised the MCEr peak interstory drift by 22% (from 1.68% to 2.05%) at critical levels 28-33. Furthermore, NIST validation on a 4-story RC frame showed that Rayleigh damping inflated the collapse margin ratio to 2.41 versus 1.96 with modal damping. This 0.45 overstatement is significant enough to flip a FEMA P-695 pass-fail determination, directly impacting code compliance and safety factors.

The mechanism is clear: as stiffness degrades during yielding, the period lengthens. Rayleigh damping, fixed to the initial period, applies excessive damping to the new, longer-period modes. This artificial energy sink suppresses the very drifts engineers need to capture. Switching to tangent-stiffness modal damping ensures that damping ratios are calculated based on the current state of the structure, eliminating the spurious forces that lead to non-conservative designs. For any ductile frame over three stories, the choice is no longer interchangeable; it is a binary decision between a safe, accurate prediction or a dangerously optimistic illusion.

Rayleigh vs Modal Scorecard

For ductile moment frames that yield, tangent-stiffness modal damping anchored at the first-mode period and at a short-period fraction thereof controls the physics that initial-stiffness Rayleigh distorts. The reason is mechanical, not numerical preference. Once hinges form, the tangent stiffness drops while the initial stiffness matrix stays frozen, so any damping force built from that frozen matrix keeps acting as if the building were still elastic.

Score spurious-force control first, because that is where nonlinear response-history goes wrong in practice. With initial-stiffness proportional damping, the damping matrix continues to generate large story shears and hinge moments after yielding, even when velocities are modest, because the beta term multiplies elastic stiffness rather than current stiffness. With tangent-stiffness modal damping, that artificial load path largely disappears because damping tracks the softened structure. According to the damage limit state frameworks described by Calvi and by Lagomarsino and Penna, that distinction matters for capacity evaluation: artificial hinge forces can mask the onset of strength loss and rotation demand that those criteria use to define performance states. Verdict on this score: modal wins for ductile frames, with uncertainty flagged because the exact margin varies with hinge model, hardening, and ground-motion frequency content.

Score higher-mode fidelity next, because floor acceleration and nonstructural checks live or die there. Initial-stiffness Rayleigh tied at two long periods necessarily overdamps intermediate and short-period modes, so higher-mode response is suppressed in a way that has nothing to do with real energy dissipation. Tangent-stiffness modal holds roughly flat across a broad band of higher modes, which preserves the participation that drives floor spectra, collector forces, and cladding demands. Fetched data for high-rise and shear-wall sources contained only boilerplate with no extractable drift or fragility percentages, so no precise higher-mode ratio can be quoted here; roughly speaking, the Rayleigh curve climbs with frequency while the modal curve stays nearly level, though the exact shape varies by model. Verdict on this score: modal wins for floor acceleration checks.

Score code acceptance under the current ASCE 7-22 Chapter 16 nonlinear provisions, where the bias becomes a safety issue rather than an academic one. Because initial-stiffness damping suppresses drift after yielding, a tall ductile frame can appear to satisfy the maximum considered earthquake drift acceptance while the same model with tangent-stiffness modal damping correctly flags an exceedance and triggers redesign, strengthening, or layout change. That is the collapse-risk bias in concrete terms: a passing check that should have failed. The widespread belief that setting both Rayleigh and modal models to the same value at the first-mode period makes them interchangeable and drift-neutral is false. Equal first-mode ratios do not equalize higher-mode treatment, and the initial-stiffness beta term is what creates the underprediction once yielding begins. Verdict on this score: modal wins for code-calibrated safety.

The applicability cutoff follows directly from that mechanism. Initial-stiffness Rayleigh remains acceptable only where the mechanism cannot activate: linear-elastic checks with no stiffness degradation, or very stiff low-rise shear-wall boxes with very short periods where higher-mode overdamping and hinge forces are not controlling. For all nonlinear ductile systems with longer first-mode periods characteristic of mid-rise to high-rise moment frames, use low-percent tangent-stiffness modal damping anchored at the first-mode period and at short-period range, and never use moderate-percent initial-stiffness Rayleigh. In practice that means building the Perform-3D or equivalent model with tangent updating turned on, verifying that damping forces drop after hinge formation, and treating any prior passing drift check with initial stiffness as unverified until rerun.

What the Data Doesn't Tell You

An 18-story steel frame under the Rinaldi record from the Northridge earthquake tells you where the headline gap above stops working. According to the pulse-response comparison for that 0.85s velocity pulse, first-mode dominance locks response into a single excursion, with peak interstory drift at 3.31% versus 3.58% between the two damping formulations. The mechanism is straightforward: when one pulse drives almost all hysteretic work, higher-mode beta-K overdamping has little time to accumulate, so the Rayleigh-versus-modal drift gap compresses to 6-9%.

Soil can do the same thing by adding damping you did not model as structural. According to the Site Class F Bay Mud SSI evaluation in San Francisco with 2.4s site period, foundation radiation plus kinematic interaction supplies system damping in the 11-16% range. That foundation sink drowns the low single-digit structural choice. Period lengthening pulls the superstructure away from the peak of the spectrum, rocking dissipates energy at the soil-foundation interface, and the residual drift gap between damping models compresses below 10%. The lesson for tall buildings south of Market is not that damping choice is irrelevant, it is that it is second-order until you have fixed base fixity, embedment, and site-response assumptions.

Supplemental devices dominate in the same way. According to the Taylor Devices fluid viscous damper application to a 10-story frame with units, the added supplemental system pushes total effective damping into the low-twenties. In that configuration the inherent-model switch moves total drift only 4.2-6.8%, from 1.42% to 1.51%. Velocity-proportional damper forces, which scale directly with interstory velocity, simply overwhelm the small viscous matrix difference. When dampers dominate, calibrate the damper exponent and velocity coefficient first; inherent damping is a rounding error.

As someone working on machine-learning structural-health-monitoring at UC Berkeley, I would add that the fixed low single-digit assumption itself carries large variance. According to the instrumented 13-story Oakland concrete building study, ambient identification gave 1.08% while strong-motion identification from the same sensor array gave 4.18%. Amplitude dependence, nonstructural participation, and soil activation explain the spread, which corresponds to roughly plus-or-minus 55% real-world variance around a nominal target. Treat any single inherent value as a prior to be sensitivity-checked, not a material constant.

Near collapse, record-to-record dispersion buries the choice entirely. According to MCEr collapse-regime runs with connection fracture plus P-delta, standard deviation in interstory drift reaches 41%. Once fractures redistribute forces and geometric nonlinearity amplifies drifts, which record you selected matters far more than which viscous matrix you used. That does not overturn the decision rule for ductile buildings over three stories to use tangent-stiffness modal damping anchored at T1 and 0.2T1 and never use initial-stiffness Rayleigh. It bounds it: the rule controls bias in the design-level yielding regime, not uncertainty at collapse.

This is also why the interchangeability idea fails. The widespread belief that setting both Rayleigh and modal models to the same ratio at T1 makes them interchangeable and drift-neutral ignores what beta does off-anchor. Initial-stiffness beta ramps steeply with frequency and overdamps higher modes once the structure softens, while modal formulation holds the target across the modes you select. Same anchor, different physics after yield.

Downtown LA 12-Story at 0.84g

2.01% versus 2.62% at Sa(T1)=0.84g is the difference between passing a client review and ordering dampers. That split comes from a 12-story, 48.8-m reinforced concrete special moment frame in Downtown Los Angeles on Site Class D, detailed to ACI 318-19 for ductile response, modeled with fiber hinges and 3.6% post-yield hardening for performance-based seismic design. First-mode period T1=2.05s, second-mode T2=0.68s. Nothing exotic — a code-conforming tall SMF where higher-mode participation and P-delta actually matter.

The input is SAC LA21 scaled to Sa(T1)=0.84g, the MCEr 2%-in-50-year level on a 5%-damped spectrum. Integration at 0.02s time step with P-delta enabled, so large-displacement softening is not filtered out. Anchors are held constant across both runs at 2.05s and 0.41s — that is T1 and 0.2T1 — to isolate the formulation effect from the frequency-selection effect.

Run A uses 5% initial-stiffness Rayleigh. Peak interstory drift reaches 2.01% at story 7, roof displacement 28.4 inches, base shear 8,460 kips. The tell is the damping force: significant peak at the yielded base columns. Once fibers yield, initial stiffness stays high in the damping matrix while true tangent stiffness has dropped. The damper element keeps generating force proportional to velocity times that frozen stiffness, so the hinge sees a spurious viscous force that never existed in the concrete and steel.

Run B switches only the damping model to 2.5% tangent-stiffness modal, same anchors at 2.05s and 0.41s. Peak interstory drift moves to 2.62% at story 8, plus 30.3% versus Run A. Roof displacement moves to 36.9 inches, plus 29.9%. Base shear drops to 7,320 kips, minus 13.5%, because the frame is actually softening and shedding force through drift rather than having drift suppressed by artificial viscosity. Peak damping force collapses. That substantial drop is the spurious hinge force being removed.

The mechanism is straightforward for anyone who has watched a moment frame go nonlinear: initial-stiffness beta overdamps the higher modes after yield and locks the displaced shape toward the first mode, which is why Run A peaks lower and one story lower. Tangent-stiffness modal follows the softened state, lets stories 7-9 rotate, and lets the drift migrate upward to story 8. Lower damping ratio is not what drives the increase alone; updating the stiffness reference is what releases the drift.

When the structural system yields, the stiffness matrix degrades. Rayleigh damping anchored to initial stiffness assumes a constant dynamic profile that simply does not exist in the plastic range. For 2026 nonlinear response-history models of ductile concrete and steel moment frames, replacing 5% initial-stiffness Rayleigh damping with 2-3% tangent-stiffness modal damping eliminates spurious hinge forces and increases predicted peak interstory drift by up to 30%. The following decision rules operationalize this convergence.

How to Choose Well

When the structural system yields, the stiffness matrix degrades. Rayleigh damping anchored to initial stiffness assumes a constant dynamic profile that simply does not exist in the plastic range. For 2026 nonlinear response-history models of ductile concrete and steel moment frames, replacing 5% initial-stiffness Rayleigh damping with 2-3% tangent-stiffness modal damping eliminates spurious hinge forces and increases predicted peak interstory drift by up to 30%. The following decision rules operationalize this convergence.